Business process automation: where to start

In short: business process automation (BPA) uses software and integrations to run repetitive work such as orders, quotes, invoices and reports without manual steps. Start with one frequent, measurable process, connect the systems you already use with tools like n8n or Make, assign an owner, then expand.

Key takeaways

Automate first the processes that are frequent, rule-based and already digital: orders, quotes, invoices, reports.
AI adoption is growing fast (Eurostat, Istat), and Bain finds that in industrial automation value is shifting towards software and data.
Most companies don't need new software. They need the tools they already have (online store, CRM, ERP, email) to talk to each other.
Without governance (an owner per workflow, error logs, controlled access) automation creates new problems instead of solving old ones.
The main obstacle is skills, not technology: training and involving the people who do the work come before choosing tools.
Run a simple sum before you start: hours saved and errors avoided against platform, setup and maintenance costs. If the payback is too long, that process is not the one to automate first.

In a growing online store or a B2B company with a sales team, a large share of the day goes into copy and paste: orders re-keyed into the ERP, quotes built from scratch, invoices checked by hand, the same report rebuilt every Monday. Each manual step costs time and adds errors: automation removes those steps and leaves people with decisions and relationships.

What is business process automation (BPA)?

Business process automation (BPA) is the use of software to run a sequence of tasks that a person currently does by hand, from the event that starts it to the final outcome. It covers the whole process rather than a single task: for example, from the order received to the invoice issued.

A few terms worth knowing:

A workflow is the sequence of automated steps. Make calls it a scenario.
A trigger is the event that starts the workflow: a new order, a submitted form, a date.
An integration connects two applications, usually through an API, the interface one program uses to exchange data with another.
RPA (robotic process automation) mimics a person clicking on a screen. It helps with legacy software that has no API, but it breaks more easily than a direct integration. There are now also AI agents that use the browser in place of a person: see the guide on what AI browser agents really automate.
Low-code / no-code platforms let you build a workflow by connecting visual blocks instead of writing code. Many are subscription services (SaaS) used through the browser, which puts automation within reach of SMEs without an IT department, although critical workflows still need someone who can design and maintain them.
An AI agent is a workflow that uses a language model to handle unstructured steps, such as reading an email or a PDF and deciding what to do with it. I cover this in more depth in the article on what AI agents change for SaaS buyers.

Why automate business processes now? What the data says

Because software and data are already inside most companies but often disconnected, and the latest figures point to a fast shift.

Across the EU, according to Eurostat, about one enterprise in five (19.95%) used AI in 2025, ranging from 17% of small enterprises to 55.03% of large ones. The most common uses were marketing and sales (34.70% of AI users) and business administration and management (31.05%).
In Italy, Istat reports that AI use among companies with at least 10 employees doubled from 8.2% in 2024 to 16.4% in 2025, reaching 53.1% of large companies.
Istat also shows how uneven the basics still are: in 2025, ERP systems were used by 48.8% of Italian SMEs and 85.9% of large companies, CRM by 21.1% of SMEs and 56.5% of large companies.

Then there is the Bain & Company figure, widely quoted and often misread. In Industrial Automation: From Control to Intelligence (April 2026), Bain expects that by the end of the decade more than 80% of the industrial automation industry's profit pools will sit at the two ends of an "hourglass": software, data platforms and AI-enabled layers on one side, accounting for more than half of the total, and smart field devices on the other, with a further 25-30%. Bain's press release adds that nearly half of the industry's revenue is expected to rely on AI-enabled offerings by 2030.

Read it carefully: it does not say that 80% of every company's profit comes from software. It says that among the vendors of automation, value is moving away from traditional control systems towards software and data. For companies that buy and use automation the message is similar: the advantage no longer sits in a single machine or application, but in how data moves between systems.

This is what many call software-defined automation: the process logic lives in software and can be changed without rebuilding the plant or replacing the ERP. In the same report Bain notes that as software decouples from hardware and interoperability improves, switching costs are falling. For a company that automates, the practical takeaway is to choose tools that connect through APIs and open formats, so you are not locked into a single vendor.

Which processes should e-commerce and B2B companies automate?

Start with processes that happen often, follow clear rules and begin with data that is already digital. Orders, quotes, invoices and reports usually top the list because they touch customers and cash and are easy to measure.

Process	How it often works today	What automation does	What to measure
Orders	Orders from the store or by email re-keyed into the ERP	Creates the order in the ERP, updates stock and notifies the customer	Order-to-dispatch time, data entry errors
Quotes	Built from scratch in Word or Excel, sent with no follow-up	Generates the quote from the price list and CRM, sends it and schedules follow-ups	Response time, acceptance rate
Invoices and payments	Due dates and payments checked by hand	Matches orders, e-invoices and payments and flags overdue accounts	Days sales outstanding, manual reminders
Reports	Data pulled by hand from several systems every week	Collects the data automatically and sends a dashboard or summary	Hours saved, report on time
Leads and CRM	Contacts from the website, ads and trade fairs scattered across files	Adds the contact to the CRM, assigns it to a rep and starts the email sequence	Time to first reply, lost leads
Support and returns	Requests sorted by hand from a shared inbox	Classifies requests, opens the ticket and answers common questions	Resolution time, requests handled without manual work
Approvals and documents	Purchase requests, contracts and expense claims circulating by email waiting for sign-off	Routes the document to the approver, logs who approved it and when, chases late approvers	Approval time, stuck documents
Stock and reordering	Stock levels checked by eye, supplier orders placed once the product has already run out	Alerts or drafts the supplier order when stock drops below a threshold	Stock-outs, rush orders
People and HR	Leave, attendance and onboarding handled with spreadsheets and email	Collects requests, updates the calendar, opens accounts and checklists for new hires	Request handling time, attendance errors

Two notes on the table. On invoicing in Italy, e-invoices go through the Exchange System (SdI) of the Agenzia delle Entrate, the Italian Revenue Agency, whose "Fatture e corrispettivi" area provides services to generate, transmit and store them and allows accrediting a web service or SFTP channel with the SdI. In practice your automation workflow rarely talks to the SdI directly: it goes through the ERP or invoicing provider you already use and takes care of linking orders, invoices and payments. For stock, shipping and returns, see the guide on SME logistics digitalization; for HR, the one on AI in HR and compliance automation.

E-commerce example

Take a typical scenario: a Shopify store that also sells to resellers. When an order comes in, a workflow records it in the ERP, checks availability, creates the shipping label and sends the tracking link. If the buyer is a reseller, it applies their price list and alerts the account manager. Every Monday a second workflow pulls sales, returns and ad spend into one report. The person who used to copy data now only checks exceptions.

B2B example

A B2B company with a field sales team receives requests from the website, trade fairs and email. A workflow brings them all into the CRM, routes them by territory and drafts a quote from the price list. The rep reviews and sends it; if there is no reply after seven days, the system reminds them to call. At month end, the report on open, won and lost quotes builds itself. To connect these steps to marketing, see the guide on choosing between an all-in-one platform and an integrated stack.

Which tools to use: n8n, Make or built-in automations?

A practical rule: use the automations already built into your software first, then an integration platform such as n8n or Make to connect different systems, and AI only where steps are unstructured.

Built-in automations. Shopify, WordPress with WooCommerce, CRMs and many ERPs include rules and notifications. They are the first layer: cheap, and nothing extra to maintain.
Integration platforms. n8n is a fair-code licensed workflow automation tool that combines AI features with business process automation; you can use it in the cloud or self-host it, for example with Docker. Make organises workflows into scenarios, with settings for how a scenario behaves when an error occurs and the option to store failed runs and resolve them later. These are the tools I use most to connect stores, CRMs, ERPs and email: n8n when volumes and data control matter, Make when speed of setup comes first. There is a dedicated guide on n8n and open-source automation for business.
AI and agents. Models such as OpenAI's help when the input has no fixed shape: an order sent as a PDF, a poorly written request, an email to classify. They belong inside a workflow with rules and checks, not on their own. A typical case is a chat assistant on the website that answers common questions and hands complex requests to a person: it works alongside the existing team instead of replacing it.

On where workflows should run: cloud platforms need no servers to manage and start immediately; self-hosting, possible for example with n8n, gives more control over data and costs at high volumes but needs someone to handle updates, backups and security.

How do you govern automation without losing control?

By deciding upfront who owns each workflow, which system is the source of truth for each piece of data, and what happens when something fails. A workflow with no owner breaks without anyone noticing.

One owner per workflow: someone knows what it does, monitors it and approves changes.
Source of truth: for each data point (customer, price, stock) one system has the final say; the others read from it.
Visible errors: every workflow logs its runs and sends an alert when it fails instead of stopping silently. In n8n, for example, you can attach an error workflow that runs when an execution fails and sends an email or a message; one error workflow can serve several workflows.
Access and credentials: service accounts with the minimum permissions needed, never an employee's personal password.
Human in the loop: decisions involving money, contracts or people stay approved by a person.
Documentation: a one-page sheet per workflow (purpose, trigger, systems, owner) so you don't depend on whoever built it.

If your workflows process personal data (individual customers, contacts, employees), the GDPR (Regulation (EU) 2016/679) applies. Two articles matter for automation. Article 28: a platform that processes data on your behalf is a processor, must provide sufficient guarantees, and the relationship must be governed by a contract or other legal act. Article 22: a person has the right not to be subject to a decision based solely on automated processing that produces legal effects on them or similarly significantly affects them, subject to the exceptions it lists. It is one more reason to keep a person in the loop when a workflow decides about customers or employees.

If AI is part of your workflows, there is also a legal framework. The EU AI Act (Regulation (EU) 2024/1689) sets risk-based rules for those who develop and those who deploy AI. It has largely applied since 2 August 2026, while AI literacy obligations were already in force from 2 February 2025. In practice: know which workflows use AI and for what, and train the people who use them. For compliance and employee data, see the article on AI in HR processes and compliance.

Why do automation projects stall? Cultural barriers

Mostly because of missing skills and the worries of the people who do the work every day, not because of technology limits. According to Eurostat, lack of relevant expertise was cited by over 70% of EU enterprises that considered AI but did not adopt it, followed by unclear legal consequences (52.52%) and data protection concerns (48.83%). Istat reports a similar picture in Italy, where missing skills held back almost 60% of the companies that assessed AI without investing.

Resistance usually has a specific cause: the person handling orders fears losing control, or their job. What works is starting from their most annoying problem, building the workflow with them and showing the result in numbers. When automation takes away the dull work and leaves the decision with the person, resistance drops quickly. Some basic training also helps: what the workflow does, how to read an error, whom to report it to.

There is also a technical obstacle that is often underestimated: getting different systems to talk to each other. Old ERPs with no API, duplicated or inconsistent data across applications, rules that exist only in the head of the person doing the job. That is why the first practical step is almost always to tidy up the process and the data, and only then build the workflow.

How do you scale automation as the company grows?

By designing workflows as small reusable pieces, tracking volumes and costs, and moving critical workflows to suitable infrastructure as they grow. A workflow that handles 50 orders a day may not handle 2,000.

Reusable modules: one "create customer in CRM" sub-workflow used by every process, instead of ten slightly different copies.
No duplicates: if an event arrives twice, the workflow must recognise it and not create two orders or two invoices.
Volumes and costs: cloud platforms charge per execution, operation or credit; above certain volumes, self-hosting is worth assessing.
Monitoring: a regular check on timings, errors and failed runs, with a named person reading it.

To decide what to keep and what to improve, you need clear indicators, the same ones used for AI projects: see the guide on measuring the value of AI projects with the right KPIs. If you sell online to other businesses, see also the guide on how B2B e-commerce works and how to launch it.

How to apply it in your company: first steps

Map three repetitive processes and, for each one, count how many times it happens in a month and how many minutes it takes each time.
Pick one with many manual steps, data that is already digital and a direct impact on customers or cash (usually orders or quotes).
Sketch the workflow on paper: trigger, systems involved, rules, exceptions, who approves. Name the source of truth for each data point.
Build it and run it in parallel for two weeks: the workflow does the work, a person checks it. Fix the exceptions.
Measure before and after (time, errors, response time) and move to the next process only when the first one is stable and has an owner.

The work I do for e-commerce and B2B companies is described in the solutions section; more guides on the topic are in the automation category.

How do you calculate the ROI of automation?

Compare the measurable benefits with the total cost of the automation, not just the platform price. The formula is simple: ROI = (benefits − costs) / costs. Alongside ROI, look at the payback period: how many months it takes for the benefits to cover the costs.

What goes into costs:

platform subscription or infrastructure;
process analysis, setup and testing;
maintenance: updates, fixes when a connected application changes, monitoring;
internal time: training and handling exceptions.

What goes into benefits:

hours of manual work removed, multiplied by the fully loaded hourly cost;
errors avoided: wrong orders, returns, invoices to correct;
effects on revenue and cash: quotes sent sooner, timely reminders, fewer lost leads.

An example with hypothetical numbers, only to show the calculation:

Item	Calculation (assumption)	Value
Manual work removed	400 orders a month × 6 minutes	40 hours a month
Monthly benefit	40 hours × €30 fully loaded hourly cost	€1,200 a month
First-year benefit	€1,200 × 12	€14,400
First-year costs	€6,000 setup + €1,200 platform + €1,800 maintenance	€9,000
First-year ROI	(14,400 − 9,000) / 9,000	60%
Payback period	€9,000 / €1,200 a month	about 7.5 months

These are example values, not market averages: for your company you need the numbers measured in the first step (how often, how many minutes). The sum leaves out errors avoided and faster replies to customers, which often matter as much as the hours. If you are looking at public funding for digital projects in Italy, see also the guide on Italian digitalization grants.

When is automating a process not worth it?

When the process is rare, unstable or made of more exceptions than rules. In those cases building and maintaining the workflow costs more than the time it saves, or the workflow keeps breaking.

It happens a few times a month: if the manual work is worth a few hours a year, payback comes too late.
It changes often: if rules, price lists or forms change every month, the workflow has to be rewritten every time.
Every case is different: if each step needs judgement and negotiation, automate only the data collection and leave the decision to the person.
The data is not digital or not reliable: digitize and clean it first, then automate.
Nobody can own it: a workflow without an owner becomes a risk, not a saving.

Mistakes to avoid

Automating a messy process: if the manual process is unclear, automation just speeds up the chaos. Simplify first, then automate.
Starting with five processes at once: none of them stabilises and the project loses credibility.
Workflows without error alerts: a workflow that stops silently is worse than no workflow.
Letting AI decide unchecked: prices, discounts and contractual replies need human approval.

Frequently asked questions (FAQ)

What is business process automation?

It is the use of software to run a sequence of repetitive tasks automatically, from the event that starts them to the final outcome, for example from order to invoice. It reduces manual work and errors and leaves people with decisions and customer relationships.

Which process should I automate first?

One that is frequent, rule-based, already digital and directly affects customers or cash. In e-commerce this is often order handling; in a B2B company, quote preparation. Measure it before and after to know whether the work paid off.

What is the difference between BPA and RPA?

BPA connects applications through integrations and APIs and automates the whole process. RPA mimics a person clicking through an application's interface: it helps with legacy software that has no API, but breaks more easily when the interface changes.

Is n8n or Make better?

It depends on volumes, data control and in-house skills. n8n can be self-hosted and suits cases where volume and confidentiality matter; Make is very visual and quick to start in the cloud. Both work for many projects: designing the process well matters more.

Does automation replace people?

In the processes described here it mainly removes repetitive tasks such as re-keying data, building reports and sending reminders. People stay on exceptions, decisions and customer relationships, which is why they should be involved from the start.

How much does it cost to automate a process?

Cost depends on how many systems need connecting, data quality and volumes. Integration platforms offer subscription plans and some, such as n8n, can be self-hosted; most of the cost still lies in analysis and configuration. Starting with one process shows you the real cost before you expand.

What is the difference between digitization and automation?

Digitizing means moving information and tasks into digital form, for example an online form instead of a paper sheet, or a CRM instead of a spreadsheet. Automating means letting software carry out the steps on that data, such as creating the order, sending the quote or chasing a payment. You digitize first and automate second: without digital, reliable data there is nothing to automate.

Sources

Use of artificial intelligence in enterprises – Eurostat (2025 data)
Imprese e ICT, anno 2025 – Istat (December 2025, in Italian)
Industrial Automation: From Control to Intelligence – Bain & Company (April 2026)
Nearly half of industrial automation revenue to rely on AI by 2030 – Bain & Company press release (7 April 2026)
AI Act: regulatory framework for AI – European Commission
n8n documentation
Scenarios – Make Help Center
Handle errors gracefully – n8n documentation
Fatture e corrispettivi – Che cos'è – Agenzia delle Entrate (in Italian)
Regulation (EU) 2016/679 (GDPR) – EUR-Lex

If you want to find out which process to automate first in your company, book a free diagnosis.